The present invention relates to the technical field of mobile communications and more particularly relates to a mobile communication system, a receiver and a method that utilize a MC-CDMA (Multi Carrier-Code Division Multiple Access) scheme and a MIMO (Multiple Input Multiple Output) scheme.

Multi carrier transmission schemes are schemes where a frequency band is segmented into multiple smaller bands (subcarriers) and signals are independently transmitted in the subcarriers. Particularly, an OFDMA (Orthogonal Frequency Division Multiple Access) scheme arranges the subcarriers in such a manner that the individual subcarriers are mutually orthogonal, resulting in higher frequency utilization efficiency for faster and higher capacity transmissions. In the OFDMA scheme, inter-subcarrier interference can be effectively suppressed, and accordingly the individual subcarriers can be used to transmit signals in parallel. As a result, the length of one symbol can be enlarged. Also, it is possible to suppress multipath interference effectively by setting a reasonably long guard interval.

In the MC-CDMA scheme, a transmission signal is code-spread over all the subcarriers. As a result, frequency diversity effect can be improved compared to a simple OFDM scheme, and thus the quality of signal transmissions can be further enhanced.

Meanwhile, in the case where the MAC-COMA scheme and the MIMO scheme are utilized together, signals are code-spread over the overall frequency band, and accordingly a reception signal received at one receive antenna corresponds to a sum signal resulting from multiplexing the spread signals over the overall frequency band corresponding to the number of transmit antennas. In the case where the simple OFDM scheme and the MIMO scheme are utilized together, a certain subcarrier signal is extracted from the sum signal, and signal detection can be performed on the extracted signal component to restore transmission signals transmitted from the respective transmit antennas. However, in the case where the MC-CDMA scheme and the MIMO scheme are utilized together, it is not significant to extract the certain subcarrier signal from the sum signal. This is because the signals are spread over the overall frequency band. If the signals can be despread, a certain signal component may be extracted from the sum signal. However, that is hard because reception signals are generally subject to frequency selective fading and accordingly the orthogonality of the spread codes may be impaired. Thus, it is necessary to simultaneously detect the signals which are spread over the overall frequency band and to which various signals are multiplexed.

As one example, it is assumed that the number of transmit antennas is N, the data modulation level is B (for example, B=4 in 16 QAM), the number of conceived multiplexed codes is P and the signal detection is conducted at the receiver side in accordance with MLD (Maximum Likelihood Detection). (See non-patent document 1 for a conventional QRM-MLD method, for example.) As stated above, the OFDM scheme can suppress the inter-subcarrier interference effectively and restrain the multipath interference within the guard interval sufficiently. In this case, the number of symbol candidates that the receiver side must consider is equal to 2^NxB. On the other hand, in the MC-CDMA scheme, the number of symbol candidates that may have to be considered for the number of multiplexed codes P is equal to 2^NxBxP. Since the number of symbol candidates increases exponentially depending on the number of multiplexed codes, the calculation amount of the signal detection significantly increases. In MIMO transmission in accordance with the MC-CDMA scheme, the signal detection can be conducted with a high accuracy, but it is hard to apply the MLD method due to its large calculation amount. On the other hand, in some signal detection methods requiring a less calculation amount such as a ZF (Zero Forcing) method and a MMSE (Minimum Mean Square Error) method, there is a risk of degrading the accuracy of the signal detection. The poor signal detection accuracy at the receiver side means that signals must be transmitted at larger power to maintain required signal quality (required SINR).

Non-patent document 1: K. J. Kim, et al., "Joint channel estimation and data detection algorithm for MIMO-OFDM systems", Proc. 36th Asilomar Conference on Signals, Systems and Computers, Nov. 2002

US 2010/027688 A1 (JP 2008 072722 A) relates to a multi input multi output orthogonal frequency division multiplexing transceiving system. A MIMO-OFDM receiver comprises a GI removal/FFT unit performing on FFT operation. The receiver furthermore comprises a QR decomposition maximum likelihood detection unit. This unit generates a complex conjugate transposed matrix of a matrix Q by QR decomposing a channel matrix corresponding to the information about channels estimated by the channel estimation unit, and generates an upper triangular matrix with respect to a channel response of vectors of received subcarriers by multiplying the complex conjugate transposed matrix by the vectors of the received subcarriers.

US 2006/179390 A1 relates to an adaptive retransmission method for frequency spreading. This document describes multicarrier CDMA.

EP 1 578 024 A1 relates to a joint detection method using minimum mean square error linear equalization (MMSI-LE) for CDMA-based mobile communication systems. The channel response matrix is made block-circulant is then row added to a square diagonal matrix representative of the effects of the MMSI variance. The modified matrix resulting from said addition is submitted to at least one of orthonormal transformation setting to zero the variance terms and making again block-circulant at the remaining part of the matrix. A transmission vector is determined by a zero forcing linear equalization method. The receiver receives the spread user symbols. The channel impulse response is estimated for each user and an impulse response matrix is created based on the estimated channels and the spreading codes. Sub-matrices of the impulse response matrix are subject to QR decompositions.

One object of the present invention is to improve the signal detection accuracy at the receiver side under the case where the MC-CDMA scheme is utilized in a MIMO based mobile communication system.

This object is accomplished by the subject-matter of the independent claims. The dependent claims describe advantageous embodiments.

In one aspect of the present invention, there is provided a receiver for use in a mobile communication system using a MC-CDMA scheme and a MIMO transmission scheme. In the mobile communication system, a set of symbols in a sequence of transmission symbols together with spreading codes for code multiplexing is converted into a code-multiplexed signal, and inverse Fourier transform is performed on the converted code-multiplexed signal, and the resulting signal is transmitted from multiple transmit antennas.

The receiver includes an extraction unit configured to perform Fourier transform on signals received at multiple receive antennas and extract signal components of individual subcarriers and a signal detection unit configured to apply a QR decomposition algorithm to the extracted signal components and estimate a transmission symbol.

The signal detection unit includes a decomposition unit configured to find a unitary matrix such that a product of a spreading matrix, a channel matrix and the unitary matrix becomes a triangular matrix, the spreading matrix having the spreading codes for the code multiplexing as matrix elements, and the channel matrix representing radio channel states between the transmit antennas and the receive antennas and an estimation unit configured to use the triangular matrix and a vector to estimate transmission symbols transmitted from the individual transmit antennas, the vector resulting from multiplication of the unitary matrix with a reception vector including the signal components received at the receive antennas.

The signal detection unit may include a candidate narrowing unit configured to provide metrics for the individual symbol candidates and narrow down the symbol candidates based on the metrics, and the metrics may represent square Euclidean distances between the symbol candidates and reception symbols on a symbol constellation.

The receiver may further include a rearrangement control unit configured to provide the decomposition unit with an indication signal for rearranging rows or columns of a matrix product of the channel matrix and the spreading matrix. The decomposition unit may find the triangular matrix and the unitary matrix in accordance with the indication signal such that the matrix product having the rearranged row or columns is equal to a product of the triangular matrix and the unitary matrix.

The rearrangement control unit may provide the indication signal such that the estimation unit estimates a symbol from a transmit antenna corresponding to greater reception power before a symbol from a transmit antenna corresponding to smaller reception power in symbol estimation in accordance with a M algorithm.

The rearrangement control unit may provide the indication signal such that the estimation unit estimates a subcarrier component of a symbol transmitted from a second transmit antenna in a first subcarrier after estimating a subcarrier component of the first subcarrier of a symbol transmitted from the first transmit antenna.

The number of surviving symbol candidates at each stage may be variably controlled depending on a radio transmission state in narrowing down the symbol candidates by the estimation unit using a M algorithm where a number of stages corresponding to a number of rows or a number of columns of the triangular matrix are utilized.

Cumulative metrics may be calculated for individual combinations of symbol candidates surviving until a certain stage and be stored as cumulative metrics for the certain stage. The estimation unit may determine a combination of symbol candidates surviving at a next stage by comparing a sum of a metric of a symbol candidate added at the next stage and the cumulative metric of a certain combination of symbol candidates with a cumulative metric of another combination of symbol candidates.

The estimation unit may calculate a total sum of the metrics for a certain combination symbols transmitted from individual transmit antennas and exclude a combination of symbols having a total sum larger than the calculated total sum from the surviving candidates.

In one aspect of the present invention, there is provided a mobile communication system including a transmitter and a receiver and using a MC-CDMA scheme ad a MIMO transmission scheme.

The transmitter includes a conversion unit configured to convert a set of symbols in a sequence of transmission symbols together with spreading codes for code multiplexing into a code-multiplexed signal, an inverse Fourier transform unit configured to perform inverse Fourier transform on the code-multiplexed set of symbols and a transmitting unit configured to transmit signals including the inverse Fourier transform applied symbols from multiple transmit antennas.

The receiver includes an extraction unit configured to perform Fourier transform on signals received at multiple receive antennas and extract signal components of individual subcarriers and a signal detection unit configured to apply a QR decomposition algorithm to the extracted signal components and estimate a transmission symbol.

The signal detection unit includes a decomposition unit configured to find a unitary matrix such that a product of a spreading matrix, a channel matrix and the unitary matrix becomes a triangular matrix, the spreading matrix having the spreading codes for the code multiplexing as matrix elements, and the channel matrix representing radio channel states between the transmit antennas and the receive antennas and an estimation unit configured to use the triangular matrix and a vector to estimate transmission symbols transmitted from the individual transmit antennas, the vector resulting from multiplication of the unitary matrix with a reception vector including the signal components received at the receive antennas.

According to the present invention, it is possible to improve the signal detection accuracy at the receiver side under the case where the MC-CDMA scheme is utilized in a MIMO based mobile communication system.

• FIG. 1 illustrates a mobile communication system for use in one embodiment of the present invention;
• FIG. 2 is a partial functional block diagram illustrating a transmitter;
• FIG. 3 is a schematic diagram illustrating calculations in a code spreading unit and a code multiplexing unit;
• FIG. 4 is a partial functional block diagram illustrating a receiver;
• FIG. 5 illustrates one example of a signal detection unit;
• FIG. 6 is a functional block diagram illustrating a receiver in detail;
• FIG. 7 illustrates different levels of difficulty of determining symbol combinations depending on the magnitude of fading correlation;
• FIG. 8 is a flowchart illustrating an operation of variation 3;
• FIG. 9 is a functional block diagram illustrating a MLD unit for use in variation 4;
• FIG. 10 illustrates an operation of variation 4;
• FIG. 11 illustrates signal point candidates ranked for reception signal points;
• FIG. 12 is a flowchart illustrating an operation of variation 4;
• FIG. 13 illustrates an operation of variation 4; and
• FIG. 14 illustrates an operation of variation 5.

• 50: cell
• 100: user equipment (UE)
• 200: base station (eNB)
• 300: access gateway
• 400: core network
• 21: code spreading unit
• 22: code multiplexing unit
• 23: inverse fast Fourier transform (IFFT) unit
• 24: guard interval addition unit
• 41: guard interval removal unit
• 42: fast Fourier transform (FFT) unit
• 44: signal detection unit
• 62: channel estimation unit
• 64: rearrangement control unit
• 210: QR decomposition unit
• 212: signal conversion unit
• 214: maximum likelihood detection unit
• 215: likelihood output unit
• 216-1-4: detection unit

For convenience, embodiments of the present invention are described using specific numerical values in order to facilitate understandings of the present invention. However, unless specifically stated otherwise, these numerical values are illustrative, and any other value may be used.

Embodiments of the present invention are described in terms of items as set forth below.

1. 1. System
2. 2. Transmitter
3. 3. Receiver
4. 4. Operations
5. 5. Details of signal detection unit
6. 6. Variation 1
7. 7. Variation 2
8. 8. Variation 2-methods 1 and 2
9. 9. Variation 3
10. 10. Variation 4
11. 11. Variation 5

FIG. 1 illustrates a mobile communication system for use in one embodiment of the present invention. In FIG. 1, a cell 50, user equipments 100₁, 100₂ and 100₃ residing within the cell 50, a base station 200, an access gateway and a core network 400 are illustrated. In one embodiment of the present invention, one or more user equipments wirelessly communicates to the base station by using both the MIMO scheme and the MC-CDMA scheme. Typically, the user equipment is a mobile station but may be a fixed station. Although the MIMO scheme and the MC-CDMA scheme are utilized in uplinks and downlinks in this mobile communication system, they may be utilized in only one of the uplinks and downlinks.

FIG. 2 illustrates an exemplary transmitter for use in a mobile communication system. In this embodiment, the transmitter is provided in a base station apparatus, but the transmitter may be provided in a base station in other embodiments. In FIG. 2, a code spreading unit 21, a code multiplexing unit 22, an inverse fast Fourier transform (IFFT) unit 23 and a guard interval addition unit (+CP) 24 are illustrated.

The code spreading unit 21 receives a series of symbol sequences to be transmitted, generates a predefined number (e.g., N_D) of copies of each symbol. The code spreading unit 21 multiplies the predefined number of symbol copies with spreading codes for code spreading. N_D represents a window size or a block size of discrete Fourier transform (and inverse transform). The symbol sequence is the symbols sequence after error correction encoding and data modulation but may be any other appropriate symbol sequence in general.

The code multiplexing unit 22 adds the different spreading codes applied symbols for individual subcarriers to generate a code-multiplexed symbol sequence.

FIG. 3 is a schematic diagram for illustrating operations in the code spreading unit 21 and the code multiplexing unit 22. The illustration is for illustrating the detailed operations, and it is not essential that individual hardware elements and functional elements corresponding to illustrated components are provided. In the illustrated embodiment, P transmission symbols d₁, ..., d_P are provided. One spreading code is provided for each of the transmission symbol. A spreading rate (spreading code length) of the spreading code is set to N_D in the illustrated embodiment. In the case where the transmitter illustrated in FIG. 2 is provided in a base station apparatus, each of the P transmission symbols d_j may correspond to data for one user. Alternatively, two or more of the P transmission symbols d_j may correspond to data for one user. In the case where the transmitter illustrated in FIG. 2 is provided in a user equipment, all the P transmission symbols correspond to data for that user equipment. In any of the cases, the code-spread transmission symbols are added (Σ) and code-multiplexed. The code-multiplexed transmission symbols are supplied to the inverse fast Fourier transform unit 23.

The inverse fast Fourier transform (IFFT) unit 23 performs inverse fast Fourier transform on the symbols associated with individual subcarriers to convert frequency domain signals into time domain signals (transmission symbols).

The guard interval addition (+CP) unit 24 adds a guard interval to the transmission symbols and supplies the resulting transmission symbols to the subsequent transmission signal generation unit (not shown). The guard interval may be generated in accordance with a cyclic prefix (CP) method.

FIG. 4 illustrates an exemplary receiver for use in a mobile communication system. In FIG. 4, a guard interval removal (-CP) unit 41, a fast Fourier transform (FFT) unit 42 and a signal detection unit 44 are illustrated.

The guard interval removal (-CP) unit 41 removes the guard interval from a baseband reception signal.

The fast Fourier transform (FFT) unit 42 performs fast Fourier transform on the reception signal to convert a time domain signal into a frequency domain signal.

The signal detection unit 44 narrows down transmission symbol candidates from signal components of individual subcarriers and finally determines the transmission symbols. As stated below, the signal detection unit 44 can not only segment the reception signal into signals from individual transmit antennas but also perform despreading. This is at least different from conventional MIMO schemes.

Operations are described below. For convenience, a transmission symbol sequence (before code spreading) transmitted from nth transmit antenna in the transmitter in FIG. 2 is written as x_n. Thus, x_n would include signals corresponding to the number of multiplexed codes. In addition, the transmission symbol x_n has N_D signal components or subcarrier components. N_D represents a window size (block size) of the discrete Fourier transform and the inverse transform. $${\mathrm{x}}_{\mathrm{n}}={\left[\begin{array}{cccc}{\mathrm{x}}_{\mathrm{n}\mathrm{1}}& {\mathrm{x}}_{\mathrm{n}\mathrm{2}}& \dots & {\mathrm{x}}_{\mathrm{nND}}\end{array}\right]}^{\mathrm{T}},$$ where T represents transposition, n is a natural number less than or equal to N_TX, and N_TX is the total number of transmit antennas.

Also, a weight factor w_i applied to i-th subcarrier at the code spreading unit 21 (i is a natural number less than or equal to N_D) is represented as follows, $${\mathrm{w}}_{\mathrm{i}}={\left[\begin{array}{cccc}{\mathrm{w}}_{\mathrm{i}\mathrm{1}}& {\mathrm{w}}_{\mathrm{i}\mathrm{2}}& \dots & {\mathrm{w}}_{\mathrm{iND}}\end{array}\right]}^{\mathrm{T}}\mathrm{.}$$

All signals simultaneously received at the receiver in FIG. 4 via N_RX receive antennas are represented as $$\mathrm{Y}={\left[\begin{array}{cccc}{\mathrm{y}}_{\mathrm{1}}& {\mathrm{y}}_{\mathrm{2}}& \dots & {\mathrm{y}}_{\mathrm{ND}}\end{array}\right]}^{\mathrm{T}},$$ where y_i represents signals for i-th subcarrier received at the individual N_RX receive antennas. $${\mathrm{y}}_{\mathrm{i}}={\left[\begin{array}{cccc}{\mathrm{y}}_{\mathrm{i}\mathrm{1}}& {\mathrm{y}}_{\mathrm{i}\mathrm{2}}& \dots & {\mathrm{y}}_{\mathrm{iNRX}}\end{array}\right]}^{\mathrm{T}}\mathrm{.}$$

In this case, the reception signal y_i for the i-th subcarrier can be written as where H_i is a channel matrix for representing radio channel states for the i-th subcarrier. where the channel matrix Hi is a N_RX×N_TX matrix, N_RX is the total number of receive antennas, and N_TX is the total number of transmit antennas. The matrix element h_i,pq of the channel matrix represents the i-th subcarrier element in a channel state (transfer function) between the pth receive antenna and the q-th transmit antenna. As one example, the matrix elements of the channel matrix may be derived from reception states of a pilot signal.

• w_i is a vector for representing a weight factor in the above spreading code, and O_ND is a 0 vector having N_D elements.
• x_n represents a signal transmitted from the n-th transmit antenna.
• N_i represents a noise component for the i-th subcarrier.

A general representation of all N_D subcarrier elements received at the respective N_RX receive antennas can be written as $$\mathrm{Y}=\mathrm{H}\times \mathrm{W}\times \mathrm{x}+\mathrm{N}$$

Such signals are supplied to the signal detection unit 44 in FIG. 4.

First, a unitary matrix Q is derived such that the channel matrix H and a weight matrix W can be represented in product form of a certain unitary matrix Q and an upper triangular matrix R. $$\mathrm{H}\times \mathrm{W}=\mathrm{Q}\times \mathrm{R}$$ where the upper triangular matrix R is M×M square matrix (M=N_D×N_TX) and consists of matrix elements r_ij (i>j) all having zeros.

In the present specification, a certain matrix A is called a unitary matrix if A^HA=AA^H=1 (identity matrix). Thus, it should be noted that the matrix A is not necessarily a square matrix. Also, the regular character "H" represents a channel matrix, and the suffix "H" represents conjugate transposition.

By multiplying Q^H to the both sides in formula (1) from the left side, the left-hand side can be written as z=Q^HY and the right-hand side can be written as $${\mathrm{Q}}^{\mathrm{H}}\times \left(\mathrm{H}\times \mathrm{W}\times \mathrm{x}+\mathrm{N}\right)={\mathrm{Q}}^{\mathrm{H}}\times \left(\mathrm{Q}\times \mathrm{R}\times \mathrm{x}+\mathrm{N}\right)=\mathrm{R}\times \mathrm{x}+{\mathrm{Q}}^{\mathrm{H}}\mathrm{N}$$

Thus, the unitary converted reception signal z can be represented as R×x, leaving noise out of consideration. $$\mathrm{z}=\mathrm{R}\times \mathrm{x}$$

Since R is an upper triangular matrix, focusing on the M-th signal component (M=N_D×N_TX), $${\mathrm{z}}_{\mathrm{M}}={\mathrm{r}}_{\mathrm{MM}}\times {\mathrm{x}}_{\mathrm{M}}$$

This means that the M-th signal component x_M can be easily estimate without consideration of signal interference from other subcarriers and/or signals from other transmit antennas.

At the first stage of the transmission symbol estimation, transmission symbol candidates are narrowed down for x_M based on formula (5). x_M must correspond to any signal point on a symbol constellation. For example, there are four possibilities (or candidates) in QPSK, and there are 64 possibilities in 64QAM. For all the possible candidates, the square Euclidean distance between the unitary converted reception signal z_M and r_MM × (x_M candidate =S_M(x)) is calculated for use as a surviving metric. $$\mathrm{e}\mathrm{1}\left(\mathrm{x}\right)={\left|{\mathrm{z}}_{\mathrm{M}}-{\mathrm{r}}_{\mathrm{MM}}\times {\mathrm{s}}_{\mathrm{M}}\left(\mathrm{x}\right)\right|}^{\mathrm{2}}$$

Among the surviving metrics, S1 (S1 ≦ C) candidates are preserved in ascending order, and the other candidates are discarded. C is equal to the total number of signal points (possible candidates) in the symbol constellation.

At the second stage of the transmission symbol estimation, the following formula based on formula (4) is used. $${\mathrm{z}}_{\mathrm{M}-\mathrm{1}}={\mathrm{r}}_{\mathrm{M}-\mathrm{1}\mathrm{M}-\mathrm{1}}\times {\mathrm{x}}_{\mathrm{M}-\mathrm{1}}+{\mathrm{r}}_{\mathrm{M}-\mathrm{1}\mathrm{M}}\times {\mathrm{x}}_{\mathrm{M}}$$ the S1 candidates derived at the first stage are used for x_M. For x_M-1, there are C different candidates. Accordingly, the surviving metrics are calculated for all possible (S1×C) combinations of x_M and x_M-1 in the above manner. $${\mathrm{e}}_{\mathrm{2}}\left({\mathrm{s}}_{\mathrm{M}}\left(\mathrm{x}\right),{\mathrm{s}}_{\mathrm{M}-\mathrm{1}}\left(\mathrm{x}\right)\right)={\left|{\mathrm{z}}_{\mathrm{M}-\mathrm{1}}-\left({\mathrm{r}}_{\mathrm{M}-\mathrm{1}\mathrm{M}-\mathrm{1}}\times {\mathrm{s}}_{\mathrm{M}-\mathrm{1}}\left(\mathrm{x}\right)+{\mathrm{r}}_{\mathrm{M}-\mathrm{1}\mathrm{M}}\times {\mathrm{s}}_{\mathrm{M}}\left(\mathrm{x}\right)\right)\right|}^{\mathrm{2}}+\mathrm{e}\mathrm{1}\left(\mathrm{x}\right)$$

The second term in the right-hand side is the surviving metric derived at the first stage. S2 (S2≦S1C) candidates are preserved in ascending order of the surviving metric e2(s_M(x), s_M-1(x)), and the other candidates are discarded.

Subsequently, the same operation is repeated. The surviving metrics are increasingly accumulated for different stages, and a combination of transmission symbols leading the minimum metric at the last stage is estimated as actually transmitted ones.

In the above description, formula (1) is calculated by using the QR decomposition and the M algorithm, but the QR decomposition and others are not essential to solve formula (1). For example, the numeral calculation may be conducted based on a MMSE (Minimum Mean Square Error) method. Note that the QR decomposition and others are preferably utilized for operation efficiency.

FIG. 5 illustrates the signal detection unit in FIG. 4 in details. The signal detection unit mainly acts to execute the above-mentioned operation. The signal detection unit includes a QR decomposition unit 210, a signal conversion unit 212, a maximum likelihood detection unit 214 and a likelihood output unit 215. The maximum likelihood detection unit 214 includes four detection units 216-1, 216-2, 216-3 and 216-4. For clarity, the four detection units are illustrated, but any number of detection units may be provided corresponding to the number of transmission signals. Each of the detection units has the same operation block, and thus the fourth detection unit 216-4 is described on behalf of the detection units 216-1 through 216-4. The detection unit includes a symbol replica generation unit 218-4, a square Euclidean distance calculation unit 220-4 and a surviving symbol candidate selection unit 222-4.

As apparently appreciated to those skilled in the art, processing elements in FIG. 5 and other blocks may be implemented in hardware, software or combinations thereof.

The QR decomposition unit 210 finds matrices Q and R such that the product of the channel matrix H and the weight matrix W can be represented in product form of the unitary matrix Q and the upper triangular matrix R (HW=QR).

The signal conversion unit 212 performs signal conversion by multiplying the conjugate transposition matrix Q^H with a vector Y having several reception signals as its components. If noise is ignored, the unitary converted reception signals can be represented in product form of the upper triangular matrix R and transmission symbols x. $$\mathrm{z}=\mathrm{Rx}$$

The maximum likelihood detection unit 214 narrows down symbol candidates of the transmission signals in accordance with the maximum likelihood detection (MLD) method. The symbol replica generation unit 218-4 in the detection unit 216-4 uses matrix elements in the upper triangular matrix R to generate the symbol candidates of the transmission signals corresponding to the reception signals x₄. The number of symbol candidates may be set to C, for example.

The square Euclidean distance calculation unit 220-4 calculates the square Euclidean distance between the unitary converted reception signals z_i and the C symbol candidates. The square Euclidean distance represents the surviving metric serving as a basis of the likelihood calculation.

The surviving symbol candidate selection unit 222-4 outputs S₁ (≦C) symbol candidates as the surviving symbol candidates based on the square Euclidean distance for the different candidates.

The likelihood output unit 215 calculates the likelihood or probability of the symbol candidates supplied from the surviving symbol candidate selection unit at the last stage. More specifically, the likelihood is represented in a log likelihood ratio (LLR). Outputs of the likelihood output unit 215 represents signal separation results, which are transmitted to the subsequent demodulation unit (e.g., turbo decoder).

In the above embodiment, the QR decomposition is applied to F=H×W. For example, the QR decomposition may be applied to a matrix G set forth below. $$G=\left[\begin{array}{c}F\\ \sqrt{N_0}I\end{array}\right]$$ where N₀ is an average noise power level measured at the receiver. I is an N_TX×N_TX identity matrix. In this case, the unitary matrix Q has N_D(N_TX+N_RX) rows and (N_D×N_TX) columns. The triangular matrix is a square matrix with (N_D×N_TX) rows and (N_D×N_TX) columns and the same dimension as the above embodiment. Although the reception signals Y is a vector with (N_D×N_RX) rows and one column in the above-mentioned embodiment, it is a vector with (N_D× (N_TX+N_RX)) rows and one column in this embodiment. Note that (N_D×N_TX) elements are zeros.

It is preferable to use the modified reception signal vector Y' and the matrix G for MMSE type QR decomposition and ZF type QR decomposition.

Also, the upper triangular matrix is utilized in the QR decomposition in the above embodiment, a lower triangular matrix can be equivalently utilized.

As stated in conjunction with the above operations, if symbols are transmitted from N_TX transmit antennas in N_d subcarriers, the reception signals Y can be written without noise as $$\mathrm{Y}=\mathrm{H}\times \mathrm{W}\times \mathrm{x}=\mathrm{F}\times \mathrm{x}$$ where H is the channel matrix, W is a spreading matrix (having spreading codes as its matrix elements) for representing weights in code spreading in the frequency direction, F is a matrix product of the channel matrix H and the spreading matrix W, and x represents transmission symbols. In the above-mentioned operation, the QR decomposition is applied to the matrix product F. Then, the transmission symbols are sequentially estimated in accordance with the M algorithm. The total number of stages in the M algorithm is equal to N_TX× N_D. The unitary matrix Q^H is multiplied with the reception signals Y, and the estimation is made from the lowest symbol sequentially. $${\mathrm{Q}}^{\mathrm{H}}\mathrm{Y}={\mathrm{Q}}^{\mathrm{H}}\times \left(\mathrm{H}\times \mathrm{W}\times \mathrm{x}\right)={\mathrm{Q}}^{\mathrm{H}}\times \left(\mathrm{Q}\times \mathrm{R}\times \mathrm{x}\right)=\mathrm{R}\times \mathrm{x}$$ where R is an upper triangular matrix with (N_DXN_TX) rows and (N_D×N_TX) columns.

Meanwhile, the product of a certain matrix A and a column vector s is equal to the product of a matrix A' resulting from column transposition for the matrix A and a vector s' resulting from the corresponding component transposition of the column vector s. $$\mathrm{A}\times \mathrm{s}=\mathrm{A}\prime \times \mathrm{s}\prime$$

For example, assuming that the matrix A consists of two rows and two columns and the column vector s is a column vector s=(s1 S2)^T with two rows and one column, the following formula holds.

In cases of higher order matrices, this relationship holds. The columns in the matrix may be rearranged in any manner. This is because the column vector elements have to be rearranged accordingly. In variation 2 of the present invention, the QR decomposition is performed by utilizing this property. $$\mathrm{Y}=\mathrm{H}\times \mathrm{W}\times \mathrm{x}=\mathrm{F}\times \mathrm{x}=\mathrm{F}\prime \times \mathrm{x}\prime$$ where the matrix F' results from some transposition of columns in the matrix F (=HXW). The column vector x' is a column vector resulting from rearrangement of elements in the column vector x corresponding to the transposition method.

In the case of the M algorithm, symbol replica candidates are sequentially narrowed down in ascending order of symbols in the transmission symbol vector for each stage. Accordingly, the order of the symbol estimation has a significant impact on the selection accuracy of the surviving symbol replica candidates. In variation 2, columns in the matrix F are rearranged to improve the selection accuracy of the surviving symbol replica candidates and thus symbol estimation accuracy.

Note that rows and columns in a matrix is a relative concept. Thus, if the transmission symbols are defined as a column vector unlike the descriptions of the present embodiment, the rows in the matrix F are rearranged. Even if the transmission symbols are defined as a row vector like the present embodiment, the generality of the present invention would not be lost.

FIG. 6 illustrates the receiver in FIG. 4 in details. In FIG. 6, a guard interval removal (-CP) unit 41, a fast Fourier transform (FFT) unit 42, a QR decomposition unit 210, a signal conversion unit 212, a MLD unit 214, a likelihood output unit 215, a channel estimation unit 62 and a rearrangement control unit 64 are illustrated.

The guard interval removal (-CP) unit 41 removes the guard interval from base band reception signals.

The fast Fourier transform (FFT) unit 42 performs fast Fourier transform on the reception signals to convert time domain signals into frequency domain signals.

The channel estimation unit 62 estimates radio channel states for individual subcarriers to derive the channel matrix H. As stated above, the channel matrices are derived for the different subcarriers. Thus, the overall channel matrix H would have (N_RX×N_D) rows and (N_TX×N_D) columns. As one example, matrix elements of the channel matrix may be derived from reception states of pilot signals.

The QR decomposition unit 210 derives the unitary matrix Q and the upper triangular matrix R based on the channel matrix H, the weight matrix W and indication signals from the rearrangement control unit 64. More specifically, the matrix F' is derived by rearranging columns in the matrix product F of the channel matrix H and the weight matrix W in accordance with the indication signals. The matrices Q and R are found such that the matrix F' can be represented in the product form of the unitary matrix Q and the upper triangular matrix R (F'=QR).

The signal conversion unit 212 performs signal conversion by multiplying the conjugate transposition matrix Q^H of the unitary matrix Q with the vector Y having several reception signals as its elements. The unitary converted reception signals can be represented in the product form of the upper triangular matrix R and the transmission symbols x, ignoring noise, $$\mathrm{z}={\mathrm{Q}}^{\mathrm{H}}\mathrm{Y}=\mathrm{Rx}\mathrm{.}$$

The MLD unit 214 narrows down symbol candidates of transmission signals in accordance with the MLD method. The MLD unit 214 calculates branch metrics for the reception signals Q^HY (=Rx) resulting from the multiplication of the unitary matrix Q^H from lower symbols sequentially. The branch metric is calculated as the square Euclidean distance between the reception signals and the symbol candidates. A predefined number (M) of symbol candidates are selected as the surviving candidates in the ascending order of cumulative branch metrics, and the operation transitions to the next stage. For each of N_TX×N_D stages, the branch metric calculation and the surviving candidate selection are conducted.

The likelihood output unit 215 calculates the likelihood of the symbol candidates supplied from the surviving symbol candidate selection unit at the last stage. More specifically, the likelihood is represented as a log (likelihood ratio (LLR). Outputs from the likelihood output unit 215 represent signal separation results and are supplied to the next stage decoding unit.

The rearrangement control unit 64 supplies indication signals to the QR decomposition unit 210. The indication signal indicates how to rearrange columns of the matrix product F of the channel matrix H and the weight matrix W. The above-mentioned formula (9) holds without depending on the specific rearrangement. The rearrangement is not uniquely determined. The rearrangement is properly determined based on any viewpoint. In this variation, the column rearrangement in the product matrix F may be determined based on reception power of transmission symbols (reception power level at the receiver) (method 1) and/or for each code spreading (subcarrier) (method 2).

The above method 1 (order control method on a per-transmit antenna basis) is described. According to the present method, the column rearrangement in the product matrix F is determined depending on the strength of reception of the transmission symbols at the receiver. In the sequential symbol estimation in accordance with the M algorithm, symbols estimated at the first through (k-1)-th stages and the k-th row in the triangular matrix R are used at the k-th stage (1≦k≦N_TXN_SF). This means that square sum (signal power) of elements between the (N_TXN_SF-k+1)-th row and the (N_TXN_SF) - th row in the (N_TXN_SF-k+1)-th column in the matrix R is used to narrow down surviving symbol replica candidates for up to k-th ordered transmission symbols. Accordingly, the estimation can be easily conducted at earlier stages (particularly the first stage) while there is a higher probability of false selection of the symbol replica candidates. According to the present method, reception signal power is measured for transmission symbols from individual transmit antennas, and the transmission symbols with greater reception signal power are estimated in higher priority. This is because high reception power decreases the false selection of symbol replica candidates. In order to achieve such priority, the order of columns in the product matrix F and elements in the transmission symbol x is controlled.

In OFDM MIMO multiplexing transmission to which the frequency domain spreading and code multiplexing according to the present method are applied, a number of transmission symbols corresponding to the number of transmit antennas are transmitted after code multiplexing. Each component of the transmission symbols transmitted from the individual transmit antennas has the same reception signal power. Thus, according to the present method, the transmit antennas providing greater reception power are identified, and symbols transmitted from the identified transmit antennas are estimated in priority over (earlier than) symbols transmitted from the other transmit antennas. Any other appropriate existing method for measuring the reception power level for the symbols from the transmit antennas may be utilized. For example, matrix elements of the channel matrix may be utilized. As stated above, the channel matrix H_i for the i-th subcarrier consists of N_RX rows and N_TX columns where N_RX is the total number of receive antennas and N_TX is the total number of transmit antennas. A matrix element h_i,pq of the channel matrix represents the channel state of the ith subcarrier component in the channel state (transmission functions) between the p-th receive antenna and the q-th transmit antenna. Accordingly, the sum of |h_i,pq|² over all the receive antenna (p=1 to N_RX) can be utilized to estimate the reception power for symbols from the q-th transmit antenna. For example, it is assumed that N_TX=N_RX=2 and N_D=3. In this case, the following formula holds for the i-th subcarrier (i=1, 2, 3). $${\mathrm{r}}_{\mathrm{i}\mathrm{1}}={\mathrm{h}}_{\mathrm{i},\mathrm{11}}{\mathrm{x}}_{\mathrm{1}}+{\mathrm{h}}_{\mathrm{i},\mathrm{12}}{\mathrm{x}}_{\mathrm{2}};$$ and $${\mathrm{r}}_{\mathrm{i}\mathrm{2}}={\mathrm{h}}_{\mathrm{i},\mathrm{21}}{\mathrm{x}}_{\mathrm{1}}+{\mathrm{h}}_{\mathrm{i},\mathrm{22}}{\mathrm{x}}_{\mathrm{2}}\mathrm{.}$$

The reception power of symbols from the first transmit antenna can be evaluated in accordance with $${\mathrm{P}}_{\mathrm{Tx}\mathrm{1}}={\left|{\mathrm{h}}_{\mathrm{i},\mathrm{11}}\right|}^{\mathrm{2}}+{\left|{\mathrm{h}}_{\mathrm{i},\mathrm{21}}\right|}^{\mathrm{2}}\mathrm{.}$$

Similarly, the reception power of symbols from the second transmit antenna can be evaluated in accordance with $${\mathrm{P}}_{\mathrm{Tx}2}={\left|{\mathrm{h}}_{\mathrm{i},12}\right|}^{\mathrm{2}}+{\left|{\mathrm{h}}_{\mathrm{i},22}\right|}^{\mathrm{2}}\mathrm{.}$$
It is assumed that the reception power of symbols from the first transmit antenna is greater than the reception power of symbols from the second transmit antenna (P_T×1>P_T×2). In the present method, the symbol x₁ = (x₁₁ x₁₂ x₁₃)^T from the first transmit antenna has to be estimated in priority over the symbol x₂ = (x₂₁ x₂₂ x₂₃)^T from the second transmit antenna. In order to implement it, columns in the product matrix F' and elements in the transmission symbol x are rearranged. Specifically, the rearranged transmission symbol x' can be written as $$\mathrm{x}\prime ={\left(\begin{array}{cccccc}{\mathrm{x}}_{\mathrm{21}}& {\mathrm{x}}_{\mathrm{22}}& {\mathrm{x}}_{\mathrm{23}}& {\mathrm{x}}_{\mathrm{11}}& {\mathrm{x}}_{\mathrm{12}}& {\mathrm{x}}_{\mathrm{13}}\end{array}\right)}^{\mathrm{T}}\mathrm{.}$$
From the viewpoint of providing the priority to the symbol from the first transmit antenna, the rearrangement is made to put the elements of the symbol x₁ into lower positions. Furthermore, the estimation order of the subcarrier elements x₁₁, x₁₂ and x₁₃ in the symbol x₁ is not uniquely determined at this stage. As one example, the estimation may be made in ascending order of the subcarrier numbers. In this case, the rearranged transmission symbol x' can be written as follows, $$\mathrm{x}\prime ={\left(\begin{array}{cccccc}{\mathrm{x}}_{23}& {\mathrm{x}}_{\mathrm{22}}& {\mathrm{x}}_{21}& {\mathrm{x}}_{13}& {\mathrm{x}}_{\mathrm{12}}& {\mathrm{x}}_{11}\end{array}\right)}^{\mathrm{T}}\mathrm{.}$$

Instead of the ascending order, any other order may be applied.

The number of antennas and the number of subcarriers as stated above are simply illustrative, and more antennas and subcarriers may be applied.

The above method 2 (order control method on a per-spreading code basis) is described. In the case where the surviving symbol replica candidates are selected in accordance with the M algorithm, there is a tendency that combinations of symbols having greater fading correlations may cause larger errors. The higher fading correlation means that symbols are subject to similar fading. In contrast, the lower fading correlation means that symbols are subject to different fading. Although not essential, the fading correlation ranges between 0 and 1, and the fading correlation closer to 1 is higher and the fading correlation closer to 0 is lower.

FIG. 7 illustrates different degrees of difficulty of the symbol combination determination depending on the magnitude of the fading correlation. For convenience, it is assumed that both the number N_TX of transmit antennas and the number N_RX of receive antennas are equal to 2. Also, symbol 1 (x₁) is transmitted from the first transmit antenna, and symbol 2 (x₂) is transmitted from the second transmit antenna. It is assumed that the symbols transmitted from the transmit antennas are data modulated in accordance with the QPSK scheme. In this case, the symbols correspond to any of predefined four signal points on a constellation. There are four possibilities for each of symbols 1 and 2, and accordingly the total number of combinations would be 16. Symbols 1 and 2 are received at the receiver in the manner where they are combined (r_i1, r_i2). As stated above, transmission signals and reception signals satisfy formulae as follows, $${\mathrm{r}}_{\mathrm{i}\mathrm{1}}={\mathrm{h}}_{\mathrm{i},\mathrm{11}}{\mathrm{x}}_{\mathrm{1}}+{\mathrm{h}}_{\mathrm{i},\mathrm{12}}{\mathrm{x}}_{\mathrm{2}};$$ and $${\mathrm{r}}_{\mathrm{i}\mathrm{2}}={\mathrm{h}}_{\mathrm{i},\mathrm{21}}{\mathrm{x}}_{\mathrm{1}}+{\mathrm{h}}_{\mathrm{i},\mathrm{22}}{\mathrm{x}}_{\mathrm{2}}\mathrm{.}$$

In the case where the fading correlation is low, each symbol is subject to significantly different fading. As illustrated in the upper-right in FIG. 7, all 16 symbol combinations can be differentiated for the combined reception signal. As a result, the symbol combination selection can be achieved with high accuracy. On the other hand, in the case where the fading correlation is high, each symbol is subject to similar fading. In the illustration, an extreme example where symbols 1 and 2 are subject to the same fading correlation is presumed. As illustrated in the lower-right in FIG. 7, the 16 symbol combinations can be partially differentiated for the combined reception signal. Only nine symbol combinations can be differentiated due to symbol overlapping. In the illustration, for signal points as illustrated in "overlapping 2 symbols" and "overlapping 4 symbols", it is impossible to determine which of the overlapping symbols is probable based on comparison of the square Euclidean distance and phase.

In the OFDM-MIMO multiplexing transmission to which the frequency domain spreading and the code multiplexing are applied, it is likely that code-multiplexed symbols transmitted from the same antenna may be susceptible to similar fading. (The fading correlation is likely to be high.)

In this variation, taking into account such circumstance, the rearrangement is made such that symbols subject to similar fading are not successively estimated. The symbol detection order is controlled such that after estimating a certain symbol, a symbol subject to fading different from the certain symbol is estimated. More specifically, the order is controlled such that after estimating a transmission symbol transmitted from a certain transmit antenna, a transmission symbol transmitted from a different transmit antenna is estimated.

The similarity determination of the fading correlation may be made in any appropriate manner. As one example, the fading correlation similarity may be determined based on similarity of the amplitude and phase of a matrix element h_i,pq in the channel matrix. For example, it is assumed that for the i-th subcarrier, pilot signals with the magnitude of 1 are transmitted from the a-th and b-th transmit antennas and are received at the pth receive antenna. In this case, the correlation between the two reception signals may be evaluated as h_i,Pa*h_i,Pb (the notation "*" represent a complex conjugate). The fading correlation similarity determination method is not limited to it, and any other method may be utilized.

The symbol detection order changed based on the fading correlation may be changed to the order where symbols from all transmit antennas are estimated for each subcarrier element. As stated above, since signals transmitted from the same antenna are subject to similar fading, they are likely to have similar fading correlation. On the other hand, since signals transmitted from different antennas are subject to different fading, they are likely to have different fading correlation. Thus, for example, the i-th subcarrier element is focused on, and N_TX transmission symbols transmitted in that subcarrier are estimated. After that, for example, (i+1)-th subcarrier element is focused on, and N_TX transmission symbols transmitted in that subcarrier are estimated. Subsequently, the same operation is performed. In this manner, symbols that tend to have different fading correlation can be sequentially estimated. In this embodiment, one subcarrier corresponds to one unit (one chip) in code spreading. Thus, the present method may be called an order control method on a per-spreading code basis. In this method, as one example, the symbol subcarrier elements are estimated in ascending order, and symbols corresponding to all the transmit antennas are estimated in each of the subcarrier elements.

For example, similar to the above example, it is assumed that N_TX=N_RX=2 and N_D=3. In this case, the rearranged transmission symbol x' can be written as follows, $$\mathrm{x}\prime ={\left(\begin{array}{cccccc}{\mathrm{x}}_{\mathrm{23}}& {\mathrm{x}}_{\mathrm{13}}& {\mathrm{x}}_{\mathrm{22}}& {\mathrm{x}}_{\mathrm{12}}& {\mathrm{x}}_{\mathrm{21}}& {\mathrm{x}}_{\mathrm{11}}\end{array}\right)}^{\mathrm{T}}\mathrm{.}$$

In this example, the subcarrier element estimation order is the ascending order. However, this is not essential. Any other order may be applied. For example, reception power levels of different subcarrier elements are compared, and the descending order of the reception power may be utilized together. For example, it is assumed that the reception power levels of the second, the third and the first subcarrier elements increase in that order. As stated in method 1 in variation 2, it is desirable that the rearrangement be made in the descending order of the reception power for improved estimation accuracy. Thus, in this case, the desirably rearranged transmission symbol x' can be written as follows, $$\mathrm{x}\prime ={\left(\begin{array}{cccccc}{\mathrm{x}}_{21}& {\mathrm{x}}_{11}& {\mathrm{x}}_{\mathrm{22}}& {\mathrm{x}}_{13}& {\mathrm{x}}_{22}& {\mathrm{x}}_{12}\end{array}\right)}^{\mathrm{T}}\mathrm{.}$$

The number of antennas and the number of subcarriers are simply illustrative, and more antennas and subcarriers may be utilized.

As described with reference to FIGS. 4 and 5, the symbol candidate combinations are narrowed down into a predefined number of combinations for each stage in accordance with the M algorithm. In the above example, the symbol candidate combinations are narrowed down into C combinations at the first stage, into S₂ (≦S₁C) combinations at the second stage, into S₃ (≦S₂C), and so on. C is equal to the number of signal points in a symbol constellation and corresponds to the total number of possible candidates (in 64QAM, C=64). S₁ corresponds to the number of surviving candidates at the first stage and is equal to 16 in the above example. The number of surviving candidates may be kept constant over all the stages or may be variably controlled. In variation 3, the number of surviving candidates is variably controlled depending on the quality of radio transmission channels.

More specific consideration is provided. Assuming Gaussian noise, the probability (likelihood function) that symbol s is transmitted from a transmitter and z is received at a receiver can be written as follows. $$\mathrm{P}\left(\mathrm{z}|\mathrm{s}\right)=\exp \left[-{\Vert \mathrm{z}-\mathrm{x}\left(\mathrm{s}\right)\Vert }^{\mathrm{2}}/\left(\mathrm{2}{\mathrm{N}}_{\mathrm{0}}\right)\right]/\sqrt{}\left(\mathrm{2}{\mathrm{\pi N}}_{\mathrm{0}}\right)=\exp \left[-\mathrm{Es}/\left(\mathrm{2}{\mathrm{N}}_{\mathrm{0}}\right)\right]$$

In this case, N₀ = σ² represents Gaussian noise power. Es represents the square Euclidean distance and is defined as Es = ||z-x(s)||². Thus, the above formula means that the symbol candidate probability (likelihood) of the square Euclidean distance Es can be represented as exp[-Es/(2N₀)]. Accordingly, the symbol candidate probability that the cumulative metric is less than or equal to (E_m,min+XN₀) at the m-th stage would be $$\exp \left[-\left({\mathrm{E}}_{\mathrm{m},\min }+{\mathrm{XN}}_{\mathrm{0}}\right)/\left(\mathrm{2}{\mathrm{N}}_{\mathrm{0}}\right)\right]/\sqrt{}\left(\mathrm{2}{\mathrm{\pi N}}_{\mathrm{0}}\right)=\exp \left[-{\mathrm{E}}_{\mathrm{m},\min }/\left(\mathrm{2}{\mathrm{N}}_{\mathrm{0}}\right)\right]\exp \left[-\left({\mathrm{XN}}_{\mathrm{0}}\right)/\left(\mathrm{2}{\mathrm{N}}_{\mathrm{0}}\right)\right]/\sqrt{}\left(\mathrm{2}{\mathrm{\pi N}}_{\mathrm{0}}\right)={\mathrm{P}}_{\mathrm{m},\max }\exp \left[-\mathrm{X}/\mathrm{2}\right],$$ where E_m,min represents a minimum value of the square Euclidean distance (distance between a reception signal point and a symbol candidate point) at the m-th stage, and ${\mathrm{P}}_{\mathrm{m},\max }=\exp \left[-{\mathrm{E}}_{\mathrm{m},\min }/\left(\mathrm{2}{\mathrm{N}}_{\mathrm{0}}\right)\right]/\sqrt{}\left(\mathrm{2}{\mathrm{\pi N}}_{\mathrm{0}}\right)\mathrm{.}$ In this manner, the number of surviving candidates can be adjusted such that symbol candidates whose cumulative metric is less than E_m,min+XN₀ can survive. As appreciated from the above formula, the probability variable X>0 is greater, the probability (likelihood function) is higher, and X>0 is smaller, the probability (likelihood function) is lower. By determining the X value appropriately, the number of surviving candidates can be appropriately controlled depending on the radio channel state. For example, for a better radio channel state, that is, for a smaller N₀, a smaller number of surviving candidates have to be selected at each stage. In the present variation 3, the value of E_m,min+XN₀ is small, and accordingly a minimum number of candidates have to be selected. On the other hand, for a poorer radio channel state, that is, for a larger N₀, a larger number of surviving candidates have to be selected at each stage. In the present variation 3, the value of E_m,min+XN₀ is large, and accordingly a larger number of candidates have to be selected.

FIG. 8 is a flowchart illustrating an operation of Variation 3. This flowchart illustrates that the cumulative metric E_m,n and the number C_m,n of symbol candidates at the m-th stage are derived from the cumulative metric E_m-1,n and the number C_m-1,n of symbol candidates at the (m-1)-th stage. At step S82, the cumulative branch metric E_m at the m-th stage is calculated from the cumulative metric E_m-i,n and symbol candidate points at (m-1)-th stage. At step S83, the minimum value E_m,min is determined in the calculated cumulative branch metrics. At step S84, the threshold E_m,min+XN₀ is calculated. It is assumed that the value of X is appropriately adjusted depending on a radio channel state. At step S85, surviving candidates are determined based on comparison of the metrics.

Considering all combinations (S_m-1×C_m-1,n) of surviving candidates S_m-1 at the previous stage and surviving candidates at the next stage in narrowing down symbol candidates in accordance with QRM-MLD method leads to significant operation load. In the present variation 4, specific examples are described for decreasing or limiting the number of candidates to be considered in the narrowing.

FIG. 9 is a partial functional block diagram illustrating relevant parts in the MLD unit of the receiver with the present variation 4. This may correspond to one of the detection units 216 in FIG. 5. Alternatively, it may correspond to the MLD unit in FIG. 6. For convenience, it is assumed that elements in FIG. 9 serve to execute operations in one detection unit in FIG. 5. Details of the operation are described in conjunction with FIG. 10. In FIG. 9, a signal point provision unit 1902, multiple square Euclidean distance calculation units 1904-1 to 1904-N, multiple addition units 1906-1 to 1906-N and a selection control unit 1908 are illustrated. The signal point provision unit 1902 includes the first to N-th provision units 1901-1 to 1901-N.

The signal point provision unit 1902 receives information regarding signal point candidates obtained at the previous stage and generates signal point candidates for use in the current stage (step). The terms "stage" and "step" may be equivalently used. For convenience, it is assumed that N=16. Each of the first to N-th provision units 1901-1 to 1901-N has the same arrangement and function, the first provision unit 1901-1 is illustratively described. The first provision unit 1901 receives one signal point candidate obtained at the previous stage and generates 16 signal points as determination targets at the current stage. The 16 signal points represent signal points in the 16QAM and correspond to white points 602 in FIG. 10. The transmitter transmits any of the 16 signal points, which is received at the receiver. The signal point received at the receiver corresponds to black point in FIG. 10, for example.

Assuming M=4 in formula (4) in the above section "4. Operations", formula (4) can be represented as follows, $${\mathrm{z}}_{\mathrm{1}}={\mathrm{r}}_{\mathrm{11}}{\mathrm{x}}_{\mathrm{1}}+{\mathrm{r}}_{\mathrm{12}}{\mathrm{x}}_{\mathrm{2}}+{\mathrm{r}}_{\mathrm{13}}{\mathrm{x}}_{\mathrm{3}}+{\mathrm{r}}_{\mathrm{14}}{\mathrm{x}}_{\mathrm{4}};$$ $${\mathrm{z}}_2={\mathrm{r}}_{22}{\mathrm{x}}_2+{\mathrm{r}}_{23}{\mathrm{x}}_3+{\mathrm{r}}_{24}{\mathrm{x}}_4;$$ $${\mathrm{z}}_{\mathrm{3}}={\mathrm{r}}_{\mathrm{33}}{\mathrm{x}}_{\mathrm{3}}+{\mathrm{r}}_{\mathrm{34}}{\mathrm{x}}_{\mathrm{4}};$$ and $${\mathrm{z}}_{\mathrm{4}}={\mathrm{r}}_{\mathrm{44}}{\mathrm{x}}_{\mathrm{4}}\mathrm{.}$$

As stated above, x₄ is first estimated based on the fourth formula. Candidates of x₄ are determined based on known r₄₄ and reception signal z₄. Then, candidates of x₃ are determined based on the third formula. Since x₃=(r₃₃)^-1(z₃-r₃₄x₄), the candidates of x₃ are estimated based on the reception signal z₃, the known value r₃₄ and a predefined number of candidates x₄ as already estimated at the previous stage. Assuming that z₃'=z₃-r₃₄x₄, this corresponds to the black point in FIG. 10. The first provision unit 1901-1 ranks 16 signal points provided at this stage based on the reception signal. FIG. 11 illustrates the ranked 16 signal points. The illustrated numbers correspond to rank order. The signal point 602 closest to z₃' is the closest (most probable and the largest likelihood), and the ranking number "1" is assigned to it. The ranking numbers may be assigned under any determination criteria. As one example, the ranking number may be assigned based on the square Euclidean distance between the reception signal and the signal points. It is preferable to define which of the signal points having the same distance has higher priority beforehand. Alternatively, the same ranking number may be assigned to the equidistant signal points.

The square Euclidean distance calculation units 1904-1 to 1904-N in FIG. 9 calculate the square Euclidean distance e_x,ix (x=1, ..., N) between signal pints supplied from the first to N-th provision units 1901-1 to 1901-N, respectively, under control by the selection control unit 1908.

For each signal point candidate, the addition units 1906-1 to 1906-N add the square Euclidean distances E_m-1,1 to E_m-1,N calculated at the previous stage to the square Euclidean distances e_1,i1 to e_N,iN, respectively, calculated at the current stage and output the results as the cumulative values of the square Euclidean distances (cumulative metrics).

The selection control unit 1908 controls operations of individual elements and main operations on the present embodiment. In general, according to the present variation, 16 signal point candidates are obtained by repeating operations including extracting one signal point among ranked signal point candidates (256 candidates in total within the signal point provision unit 1902) in the probability order, calculating the square Euclidean distance, updating comparison amounts, extracting the next one signal point and calculating the square Euclidean distance. The square Euclidean distance has to be calculated only 16 times, which significantly reduces the operation load for signal separation.

FIG. 12 is a flowchart illustrating an operation according to the present variation. The flow starts with step 2002 and proceeds to step 2004. At step 2004, parameters for use in calculation are initialized. Parameter j is a counter for indicating one of all 16 signal point candidates supplied at the current stage and is set as j=1 (1≦j≦N=16). E_m-1,1 to E_m-1,N are cumulative values or cumulative metrics of the square Euclidean distance calculated at the (m-1)-th stage (previous stage), and individual cumulative metrics such as E_m-1,1 are also referred to as cumulative branch metrics. The maximum cumulative branch metrics M_m,1 to M_m,N are used as criteria of magnitude comparison as stated below and are set to the cumulative branch metrics E_m1,1 to E_m-1,N, respectively, obtained at the previous stage (M_m,1=E_m-1,1 ..., M_m,N=E_m-1,N). In this example, N=16. The ranking numbers i₁ to iN indicate the order or likelihood for the 16 signal point candidates in the first provision unit 1901-1 and are set to one.

At step 2006, the magnitude of the 16 maximum branch metrics M_m,1 to M_m,16 are compared with each other, and the minimum (M_m,x) is selected. The x identifies the number of the selected signal point candidate in the signal point candidates at the previous stage. The signal point candidate associated with the maximum branch metric M_m,x is obtained from the x-th provision unit 1901-x.

At step 2008, the signal point candidate identified with the ranking number i_x is selected among the signal point candidates associated with the maximum branch metric M_m,x as the j-th signal point candidate at the current stage.

At step 2010, the square Euclidean distance (e_x,ix) is calculated for the selected signal point candidate. The cumulative branch metric E_m,j is also calculated in accordance with E_{m, j} = e_{x, ix} + E_{m-1, x}.

At step 2012, various parameters are updated. The maximum cumulative branch metric M_{m, x} is updated with E_{m, j}. The ranking number i_x is incremented by one. The counter j is also incremented.

At step 2014, it is determined whether the counter j is less than or equal to the maximum value N, and if the counter j is less than or equal to N, the flow returns to step 2006 where the same procedure is repeated, and otherwise, the flow proceeds to step 2016 and ends.

When the operation at one stage ends, the next stage is performed, and the same procedure is repeated to the last stage. In a signal separation operation, the cumulative metrics for the finally surviving 16 signal point candidates are compared to each other, and the signal point associated with the minimum cumulative metric (cumulative value of the square Euclidean distance) is determined as a reception signal point.

FIG. 13 illustrates an exemplary operation for j=1 to 5 in the flow chart in FIG. 12. In other words, FIG. 13 schematically illustrates specific operations for selecting five signal point candidates among 256 (=16 ×16) signal point candidates from the signal point provision unit 1902.

Initially, it is assumed that x=1 for j=1. In this case, the signal point candidate identified with the ranking number i₁=1 (initial value) is selected among the signal point candidates supplied from the first provision unit 1901-1. The square Euclidean distance e_1,1 is calculated for the selected signal point candidate and added to the cumulative value E₁ derived at the previous stage (Although it is E_m-1,1, the suffix "m-1" is omitted for clarify). The maximum cumulative branch metric M_m,1 is updated from E_m-1,1 to e_1,1+E_m-1,1. The ranking number i₁ is incremented by one (i₁=2).

For j=2, at step 2006, magnitude comparison is conducted for the updated M_m,1 and the others M_m,2 to M_m,16, and it is assumed that M_m,1 is the minimum again. Thus, x=. In this case, the signal point candidate identified with the ranking number i₁=2 is selected among the signal point candidates supplied from the first provision unit 1901-1. The square Euclidean distance e_1,2 is calculated for the selected signal point candidate and added to the cumulative value E₁ derived at the previous stage. The maximum cumulative branch metric M_m,1 is updated from e_1,1+E_m-1,1 to e_1,2+E_m-1,1. The ranking number i₁ is incremented by one (i_i=3).

For j=3, at step 2006, magnitude comparison is conducted for the updated M_m,1 and the others M_m,2 to M_m,16, and it is assumed that M_m,2 is the minimum. Thus, x=2. In this case, the signal point candidate identified with the ranking number i₂=1 (initial value) is selected among the signal point candidates supplied from the second provision unit 1901-2. The square Euclidean distance e₂, is calculated for the selected signal point candidate and added to the cumulative value E₂ derived at the previous stage. The maximum cumulative branch metric M_m,2 is updated from E_m-1,2 to e_2,1+F_m-1,2. The ranking number i₂ is incremented by one (i₂=2).

For j=4, at step 2006, magnitude comparison is conducted for the updated M_m,1 and M_m,2 and the others M_m,3 to M_m,16, and it is assumed that M_m,1 is the minimum. Thus, x=1. In this case, the signal point candidate identified with the ranking number i₁=3 is selected among the signal point candidates supplied from the first provision unit 1901-1. The square Euclidean distance e_1,3 is calculated for the selected signal point candidate and added to the cumulative value E₁ derived at the previous stage. The maximum cumulative branch metric M_m,1 is updated from e_1,2+E_m-1,1 to e_1,3+E_m-1,1. The ranking number i₁ is incremented by one (i₁=4).

For j=5, at step 2006, magnitude comparison is conducted for the updated M_m,1 and M_m,2 and the others M_m,3 to M_m,16, and it is assumed that M_m,2 is the minimum. Thus, x=2. In this case, the signal point candidate identified with the ranking number i₂₌₂ is selected among the signal point candidates supplied from the second provision unit 1901-2. The square Euclidean distance e_2,2 is calculated for the selected signal point candidate and added to the cumulative value E₂ derived at the previous stage. The maximum cumulative branch metric M_m,2 is updated from e_2,1+E_m-1.2 to e_2,2+E_m-1,2. The ranking number i₂ is incremented by one (i₂=3).

In this manner, the five signal points and the cumulative branch metrics are calculated, and the candidates (1) - (5) survive. Actually, the same procedure is repeated until 16 signal point candidates are selected.

As stated above, when the symbol candidates are narrowed down in accordance with the QRM-MLD scheme, considering all combinations (S_m-1×C_m-1,n) of the surviving candidates S_m-1 at the previous stage and possible candidates at the next stage increases operation load. Variation 5 described below decreases or limits the number of candidates in a manner different from variation 4.

For convenience, as in the above example, it is assumed that four symbols x₁, x₂, x₃ and x₄ are transmitted and received at a receiver. the 16QAM is assumed as the data modulation scheme in variation 4 while the BPSK scheme is assumed in variation 5 for clarity. In other words, each symbol x_i (i=1, 2, 3, 4) corresponds to a signal point "-1" or "1". The relationship between transmission signals and reception signals is represented as follows, $${\mathrm{z}}_{\mathrm{1}}={\mathrm{r}}_{\mathrm{11}}{\mathrm{x}}_{\mathrm{1}}+{\mathrm{r}}_{\mathrm{12}}{\mathrm{x}}_{\mathrm{2}}+{\mathrm{r}}_{\mathrm{13}}{\mathrm{x}}_{\mathrm{3}}+{\mathrm{r}}_{\mathrm{14}}{\mathrm{x}}_{\mathrm{4}};$$ $${\mathrm{z}}_2={\mathrm{r}}_{22}{\mathrm{x}}_2+{\mathrm{r}}_{23}{\mathrm{x}}_3+{\mathrm{r}}_{24}{\mathrm{x}}_4;$$ $${\mathrm{z}}_{\mathrm{3}}={\mathrm{r}}_{\mathrm{33}}{\mathrm{x}}_{\mathrm{3}}+{\mathrm{r}}_{\mathrm{34}}{\mathrm{x}}_{\mathrm{4}};$$ and $${\mathrm{z}}_{\mathrm{4}}={\mathrm{r}}_{\mathrm{44}}{\mathrm{x}}_{\mathrm{4}},$$ which correspond to formula (4) in the section "4. Operations" in the case of M=4.

The method according to variation 5 is described with reference to FIG. 14. Each symbol x_i (i=1, 2, 3, 4) corresponds to "-1" or "1". In the illustration, the metric is represented as e₁₁ for x₄ of "-1", and the metric is represented as e₁₂ for x₄ of "1". As stated above, the metric is an amount calculated as the square Euclidean distance between a predefined signal point on a constellation and a reception signal point. In addition, the metric is represented as e₂₁ for x₄ of "-1" and x₃ of "-1". The metric is represented as e₂₂ for x₄ of "-1" and x₃ of "1". The e₂₁ and e₂₂ are cumulative metrics of the metric associated with x₄ and the metric associated with x₃. Similarly, the cumulative metrics are represented as e₃₁, e₃₂, e₃₃, e₃₄, ..., e₄₁, e₄₂, ... for combinations of symbol candidates for x₄, x₃, x₂ and x₁. Note that all the cumulative metrics are not calculated. First, the cumulative metric is calculated for one combination of certain candidates for the four transmission symbols x₄, x₃, x₂ and x₁. For example, it is assumed that x₄="-1", x₃="-1", x₂="-1" and x₁="-1". Then, the cumulative metric e₄₁ is calculated. The cumulative metric e₄₂ is calculated for the case of x₄="-1", x₃="-1", x₂₌="-1" and x₁="-1" and is compared to the cumulative metric e₄₁. A smaller one of the cumulative metrics is probable (higher (likelihood). Thus, the combination of symbol candidates leading to the minimum cumulative metric in the bottom line e₄₁, e₄₂, ... would be the most probable. For this reason, it may be theoretically possible that all the cumulative metrics e₄₁, e₄₂, ... in the bottom line are calculated, the minimum cumulative metric is found through the magnitude comparison for the calculated metrics and the optimal symbol candidate combination is identified. However, this has heavy operation load.

Meanwhile, the cumulative metric is calculated through the addition of the square Euclidean distance and accordingly must have increase property. For example, in FIG. 14, the metrics e₁₁, e₂₁, e₃₁ and e₄₁ at the first, second, third and fourth stages are larger in that order. It is assumed that as a result of comparison between the cumulative metric e₄₁ at the fourth stage and the cumulative metric e₂₂ at the second stage, the cumulative metric e₂₂ is larger than e₄₁. In this case, e₃₃ and e₃₄ calculated at the third stage corresponding to the cumulative metric e₂₂ would be larger than the cumulative metric e₂₂. Similarly, e₄₅, e₄₆, e₄₇ and e₄₈ are further larger. Accordingly, according to variation 5, if the cumulative metric e₂₂ is larger than e₄₁ as a result of comparison between the cumulative metric e₄₁ at the fourth stage and the cumulative metric e₂₂ at the second stage, symbol combinations subsequent to e₂₂ are not considered. In other words, the cumulative metrics e₃₃, e₃₄, e₄₃, e₄₆, e₄₇ and e₄₈ are not calculated. This is because these would be larger than the cumulative metric e₄₁. In this manner, in variation 5, the cumulative metric e₄₁ corresponding to the certain symbol candidate combination is calculated, and combinations having the cumulative metrics, such as e₂₂, larger than the calculated cumulative metric are excluded from the surviving candidates. This can decrease the number of calculations of the cumulative metrics. In addition, combinations having the cumulative metrics, such as e₂₂, larger than a predefined constant R may be excluded from the surviving candidates. By setting the R to a small value expected to be larger than the finally found minimum cumulative metric, it is possible to further reduce the number of calculations of the cumulative metrics.

The present invention has been described with reference to the specific embodiments, but the embodiments are simply illustrative, and those skilled in the art will understand various variations, modifications, alterations and substitutions. In the above description, some specific numerical values are used for better understanding of the present invention. Unless specifically indicated, however, these numerical values are simply illustrative and any other suitable values may be used. For convenience of explanation, apparatuses according to the embodiments of the present invention have been described with reference to functional block diagrams, but these apparatuses may be implemented in hardware, software or combinations thereof.